Operate a Cell-Free Biofoundry using Large Language Models

Abstract

In this paper, we present a novel approach to optimizing cell-free protein synthesis (CFPS) systems using artificial intelligence (AI), specifically leveraging ChatGPT-4 for code generation and active learning (AL). This study aims to automate and enhance the process of producing antimicrobial proteins, namely colicin M and colicin E1, in CFPS systems. We developed an automated workflow that employs an iterative Design-Build-Test-Learn (DBTL) cycle, integrating a newly implemented AL method with cluster margin (CM) selection to efficiently explore experimental conditions. The workflow components, including modules for sampling, plate design, instruction generation, and data analysis, were coded using ChatGPT-4 without further human modification. By employing this automated approach, significant improvements in protein yields were achieved, with a 9-fold increase for colicin M and a 3-fold increase for colicin E1 compared to standard buffer compositions. The use of LLMs in conjunction with AL demonstrated the potential of AI-driven methodologies to accelerate the optimization of complex biological processes and reduce manual intervention. The study also discusses limitations such as variability in CFPS and suggests future improvements in automation, reproducibility, and integration of diverse liquid handling systems to further enhance the scalability and efficiency of cell-free biofoundries.